This post was partly inspired by Sean Kilgore’s recent blog about Common Errors in the Paved Path metaphor. Sean’s language mirrors a metaphor that surfaced within my team about our On-Call Rotation. Sean contrasts the “paved path” an organization provides to its developers with the “vehicles” each team puts on that path. My team talks about being on-call in terms of “taking the wheel of the vehicle,” instead of the more common “being responsible for fixing anything that goes wrong.”
Taking a Turn at the Wheel
Most of us have taken a road-trip with friends at some point in our lives. On such trips the group usually shared the driving duties. Everyone took a turn at the wheel.
Taking a turn at the wheel came with certain responsibilites:
- Understanding and obeying the rules of the road.
- Protecting the life and safety of the occupants and cargo.
- Protecting the vehicle and its ability to operate.
It might also have come with some privileges:
- Control of the air conditioning.
- Control of the music.
- Permission to wake people, if necessary, to protect the occupants or vehicle.
It certainly did not come with the right to:
- Have the vehicle repainted, re-upholstered, or upgraded to “our standards.”
- Sell the vehicle or trade it for one we thought was “better.”
- Force everyone to take another mode of transportation because “I don’t feel like driving right now.”
Put another way, the driver does not become the owner of the car. They also do not become the mechanic, the tow truck, or the electrician.
If the engine was overheating we expected the driver to lower speed, ask everyone to open their windows, and crank up the heater. We did not expected them to pull over and fix the radiator or perform an oil change on the side of the road.
When a tire was punctured, the driver was responsible for pulling over. They may or may not have been the one who changed the actual tire – but at that point, they were not driving.
There was no expectation that the driver be (or become) an expert on engines or drivetrains or the car’s electrics. They were just expected to be awake, present, and commited to our safe arrival a the destination.
Being On-Call is Taking a Turn at the Wheel
Being the Driver of a car is a great metaphor for being on-call. And here we are talking about being the person who “carries the pager.” If something alerts, it will be your phone that rings. You are responsible for:
- Acknolwedging the page and investigating the alert.
- Assessing the impact of the alert on the safety of the system or the business’s ability to operate.
- Taking any action necessary to preserve the safety of the system and the business.
That last item is often misinterpreted. The responsibility to take action implies the right to run commands on production systems. On-call can restart servers, kill pods, or manipulate the database. But it also authorizes them to “wake someone up.” As the Driver, on-call might know how to change a tire – and that is great – but if they can’t, they are expected to pull over and ask for help.
A driver who continues to drive on a “flat” tire is not doing anyone any favors.
Who is Not On-Call?
There can be push-back to the idea that on-call is allowed to wake up other people. This often comes in the form of the question,“Wait, doesn’t that mean I’m always on call?”
To answer this, I defer to Charity Majors’ popular ON CALL SHOULDN’T SUCK: A GUIDE FOR MANAGERS:
As for engineers who write code for 24×7 highly available services, it is a core part of their job is to support those services in production. (There are plenty of software jobs that do not involve building highly available services, for those who are offended by this.) Tossing it off to ops after tests pass is nothing but a thinly veiled form of engineering classism, and you can’t build high-performing systems by breaking up your feedback loops this way.
So the answer is: yes. Engineers are always on-call for their service.
This can raise the question,“If that’s the case, then why doesn’t each team have it’s own on-call? And why can’t the alerts route directly to them? Why have a shared on-call at all?”
Let’s apply that question to our metaphor: Would you accept a condition where your “engine is overheating” light pings your mechanic instead of you? Would it become your mechanic’s job to phone you or chase your car around to get you to slow down or stop? That sounds like a lot of engines are going to be driven into the ground by people who were never told there was an issue.
The fact is, many alerts can be mitigated without in-depth knowledge of the application. That is the purpose of runbooks: to provide instructions on mitigating, not fixing, problems.
I do support the idea of individual teams having an on-call rotation for escalations. If the org is large enough, and has enough resources, then everyone “can take their own car,” and have their own alerts and rotations. But many engineering orgs simply do not have the resources to support this. The idea I want to put forward is that org-wide on-call rotations can succeed if management clearly frames the goal of on-call when engineers join the rotation.
The Goal of On-Call
I’ve heard that one motivation for org-wide on-call is “so everyone, eventually, learns the whole application.” This framing can lead to conflict and frustration, as members of one team helpfully offer constructive criticism to members of another in the name of “making on-call suck less.” It all but guarantees that people step on one another’s toes. It fosters an “us vs them” mentality, and usually makes on-call (and the whole org) suck more.
The goal of org-wide on-call is to maintain global awareness of how the app behaves from the perspective of the customer. Even Charity Majors recommends that managers put themselves in the on-call rotation. This is not to make them full-stack engineers. This is because managers, also, need to be thinking about the customer impact of changes moving through the pipeline.
We build applications to serve customers. That’s the difference between being in business and being in a hackathon. No engineer is made better at their job by ignoring the metrics that impact the customer’s experience. No code is better for being less observable, no architecture is better for being impossible to troubleshoot (or impossible to manipulate at runtime).
It is not reasonable, and it is not safe, to constantly treat the holistic health of the app as “someone else’s job.” It is not reasonable, and it is not safe to force a small group of people do all the driving. In the name of safety, and fostering common purpose, many orgs opt for the org-wide rotation, with everyone taking a turn at the wheel.
Of course, this re-framing is only half the job. If Engineers are going to be “driving” during on-call week, that means they’re not going to be doing anything else. Going back to our metaphor, imagine being asked to take your turn at the wheel, and then the person handing you the keys says:
“Oh – and since you’re driving, you should also be able to tune the carburetor and rotate the tires. Because you’re driving anyway.”
Doesn’t make a lot of sense, does it?
Loading up on-call engineers with sprint tickets forces them to switch context when an alert comes in. This can cost them, and the org, up to 40% of their productivity.
If these engineer cares about their performance reviews, that could mean they put in 40% more time and effort to catch up. This leads to frustration, conflict, and burnout. On the other end of the spectrum, they could truly divide their efforts and attention. With all the context switching, that means their remaining 60% of effort is split between on-call and sprint work. That’s 30% effort on either task.
Giving on-call engineers sprint work is asking for less than 1/3 of an engineer on that sprint work, and less than 1/3 of an engineer on-call.
That is not reasonable, and it’s not safe.
Framing On-Call in terms of Engineers “Taking a Turn at the Wheel” helps both Management and Engineering maintain a clear understanding of the boundaries that make these org-wide rotations successful.